A quick peek into my Stable Diffusion workflow. I’m working on creating a couple of custom checkpoint merges, and creating a few LoRAs for backgrounds/settings. So I’ve been creating these massive contact sheets to evaluate the different checkpoints against eachother to see how they handle the same prompt at different steps.
I’ll award an emoji gold star to anyone that can identify the game the background lora was trained on. Extra points for the specific location.
Warning: Extra large X/Y comparison grid ahead. Each row is a different checkpoint merge. Let me know if you prefer any checkpoint over another. I’m trying to weed out anime faces and get more natural photograph, fabric folds, and of course hands and face variety, while keeping things soft and pretty: https://files.catbox.moe/1k6mmr.jpg
There is a glacial stream running behind the buildings in Colter, which was captured in the training images I took. I think thats where the water is coming from. Of course, SD is embellishing by adding yachts in some cases, lol.